QuantCrystal · Pro tier
The algorithmic layer of MarketCrystal.
MarketCrystal is where Mark reads the market in plain language, free. QuantCrystal is where the engine underneath it — Trismegistus — is put to work: simulating strategies, backtesting them through a real execution path, and keeping a scored record of what it predicted.
Free access at launch — sign up now.
We're taking the first 100 users. They get the workbench free when it opens — simulation, backtests, and the datasets as each one lands.
Nothing here is a track record. No performance figures are published on this site because none have been earned in public yet.
What you get
Four things, done without flattery.
Monte Carlo simulation
Your strategy's equity curve is one draw from a distribution. We resample it in blocks to show the other draws — the outcome spread, the drawdown percentiles, and the probability the account is dead before the edge shows up.
Read more →Backtesting through a real engine
No vectorized approximations of fills. Signals go through the same execution engine that would trade them, with the same order types, the same costs, and parameters frozen before the test window opens.
Read more →Curated datasets
Macro series are published now, and a survivorship-free equity universe including delisted names is mirrored internally. Full-market bars, crypto ticks, and filings join the catalog as licensing clears. The point of all of it: a backtest that is not quietly reading a universe containing only the survivors.
Read more →Bot-building guidance
Advisory, not a signal service. Help turning a hypothesis into a specified, testable, risk-bounded system — and an honest read on when the answer is that the edge is not there.
Read more →The engine
Trismegistus
Trismegistus is the analysis engine behind both sites. On marketcrystal.ai it speaks through Mark, in prose, for free. Here it is exposed as machinery: the regime classifier, the nowcasts, and the ledger that keeps score.
Markets do not have one behaviour. The engine classifies the current environment rather than assuming a single stationary process, because a strategy's expectancy is conditional on the regime it runs in.
Estimates of present-state conditions from data that is already observable, rather than forecasts of distant outcomes. The near term is where the signal-to-noise is least hostile.
Every prediction is written down before the outcome, with its resolution criteria, and scored when it resolves — hits and misses alike. A forecast that is never graded is marketing, not research.
Architecture
The engine runs on Cloudflare. A Workers API owns simulation, backtesting, and ledger resolution, with D1 holding run records and results; this site is the static front end against it. Everything is deployed at the edge rather than behind a single regional host, so a backtest is dispatched from wherever the request lands.
The API answers today under internal keys. The public endpoint opens at launch, on its own domain and with published rate limits — the specifics land on this page when the door opens, not as a promise before it.
Pricing
Two paid tiers.
The reading of the market is free on MarketCrystal. What the paid tiers cover is the tooling that consumes compute and licensed data — simulation runs, backtest jobs, dataset access, and engine output.
Price boards, Mark's written analysis, and the newsletter. Lives at marketcrystal.ai and is free — it is not a trial of this, and it is not time-limited.
- Cost-per-unit price boards
- Mark's market commentary
- Newsletter
- No account required
The simulation and backtesting workbench. For traders who want to test their own systems properly rather than consume someone else's calls.
- Monte Carlo simulation tools
- Backtest reports through the real execution engine
- Dataset access as each set is released
- Full methodology documentation
Everything in Pro, plus the engine's own output — what Trismegistus classifies, forecasts, and gets scored on.
- Everything in Pro
- Trismegistus signals and regime calls
- Resolution-tracked prediction ledger
- Engine API access when the public endpoint opens
- Bot-building advisory
The first 100 pay nothing
Sign up now and you get the workbench free when it opens — no card, no checkout, no trial clock. The prices above are what the tiers cost after that, in USD per month. Billing goes live at launch and the full terms are published before any charge is ever made.
Feature lists describe what each tier includes at launch. Where something arrives later than opening day, its line item says so.
Privacy
What is yours, and what is public.
Every account is private by default. Publishing a record is a per-account choice, and it exposes exactly two things about you plus the shape of your performance. This is the whole model — there is no other tier of sharing.
| Private always | Public if you opt in | |
|---|---|---|
| Email, display name, company name | Yes | Never |
| Handle and avatar | — | Yes |
| Performance trend normalized curve, stats, days live, trade count | — | Yes |
| Positions, symbols, orders, dollar amounts, the algorithm | Yes | Never |
Why the curve cannot leak your account size
A published curve is a percent series anchored at zero on its first day. The dollars never leave the server: the queries that build a public profile name their columns explicitly and have no access to a cash figure. You publish the shape of a track record without publishing the size of the book behind it.
Nothing published is self-reported either. Public figures are derived from an append-only fills journal written by the fill engine, which is why they can be audited and why a leaderboard place has to be earned over 30 days and 20 filled orders.
Free at launch
Free access at launch — sign up now.
We're taking the first 100 users. One message when access opens, and it carries your login. No drip sequence, no urgency theatre, no resale of your address.